|Publication number||US7203338 B2|
|Application number||US 10/538,483|
|Publication date||Apr 10, 2007|
|Filing date||Dec 11, 2002|
|Priority date||Dec 11, 2002|
|Also published as||US7466844, US20060062429, US20070086623|
|Publication number||10538483, 538483, PCT/2002/39619, PCT/US/2/039619, PCT/US/2/39619, PCT/US/2002/039619, PCT/US/2002/39619, PCT/US2/039619, PCT/US2/39619, PCT/US2002/039619, PCT/US2002/39619, PCT/US2002039619, PCT/US200239619, PCT/US2039619, PCT/US239619, US 7203338 B2, US 7203338B2, US-B2-7203338, US7203338 B2, US7203338B2|
|Inventors||Arun Ramaswamy, Daniel J. Nelson, Venugopal Srinivasan|
|Original Assignee||Nielsen Media Research, Inc.|
|Export Citation||BiBTeX, EndNote, RefMan|
|Patent Citations (55), Non-Patent Citations (15), Referenced by (27), Classifications (14), Legal Events (6)|
|External Links: USPTO, USPTO Assignment, Espacenet|
This patent arises from a continuation of International Patent Application Ser. No. PCT/US02/39619, which was filed on Dec. 11, 2002.
This disclosure relates generally to image analysis and, more particularly, to methods and apparatus to count people appearing in an image.
Audience measurement of broadcasted television and/or radio programs has been practiced for many years. Audience measurement devices typically collect two kinds of information from households, namely, tuning information (e.g., information indicating the content presented to the audience such as channel information, time of consumption information, program information, etc.) and people information (e.g., information about the demographics of the audience). These two types of information are combined to produce meaningful ratings data.
People information has historically been gathered by people meters. People meters have been constructed in many different manners. For example, some people meters are active devices which seek to determine the composition of the audience by, for instance, analyzing visual images of the audience to actively determine the identity of the people in the audience. Such active determination involves comparing facial features of an individual appearing in a captured image to one or more previously stored facial feature images to search for a match. Other people meters are passive devices which prompt the members of the viewing audience to identify themselves by logging themselves in at specific times. These specific prompting times can be independent of the tuning information and at fixed time intervals (i.e., time-based prompting) or they can be tied to the tuning information and be performed, for example, when the channel changes (i.e., channel change-based prompting).
The time-based prompting technique poses a danger of under sampling or over sampling the data. For example, if the prompts are spaced too far apart in time, audience members may enter or leave the room between prompts. If the audience does not notify the people meter of such entrances/exits, audience composition data and audience change timing is lost. Alternatively, if the time prompts are spaced too closely in time, the audience members may become annoyed and/or reduce their compliance with the prompt requests. Again, audience composition data is lost in such circumstances.
The channel change-based prompting technique discussed above poses the danger of over sampling the data. As explained above, such overly frequent prompting may cause irritation and/or result in a decrease in compliance and a corresponding loss of data collection and/or invalid data.
As shown in
Alternatively or additionally, in the visual presentation context (e.g., television viewing), codes embedded in the vertical blanking interval of the program being viewed may be utilized by the content collector 14 to positively identify the program being consumed by the audience.
A detailed illustration of an example implementation of the apparatus 10 is shown in
For the purpose of determining a number of people appearing in the images captured by the image sensor 18, the audience change detector 12 of the apparatus 10 is further provided with a people counter 20. The people counter 20 may determine the number of people within the image(s) in many different ways. However, a preferred method identifies people within the image(s) by detecting changes indicative of movement between successive images. An example people counter 20 and an example manner of implementing the same are discussed below in connection with
In order to determine if the number of audience members has changed, the audience change detector 12 is further provided with a change detector 22. The change detector 22 compares the number of people counted in the image(s) by the people counter 20 to a value representative of a previous number of people in the audience. The value representative of the previous audience count may, for example, be the audience count the people counter 20 developed in analyzing the last image or set of images, or, in, for example, the case of the first audience image analysis (e.g., the first image(s) collected after a power-up event), it may be a default value (e.g., 0). If a difference exists between the audience count developed by the people counter 20 and the previous number of people in the audience, the change detector 22 develops an output signal indicating an audience composition change has been detected.
As shown in
Regardless of the type of signal employed (e.g., visual, audible, etc.), in the illustrated example the people counter 20, the change detector 22 and the prompter 24 cooperate to prompt the audience member(s) to log themselves(s) in whenever a change in the number of audience members occurs. As a result, the audience is neither oversampled (i.e., prompted excessively), nor undersampled (i.e., prompted too infrequently such that audience change times are missed). Also, in the event all audience members leave the room, the apparatus 10 automatically detects and records that there is no audience members, thereby collecting accurate audience measurement data even when no audience member is present to respond to a prompt
In order to receive data from the audience member(s), the audience change detector 12 is further provided with an input device 26 such as a conventional IR transmit-receive pair, a mouse, a keyboard, a touchscreen, a touchpad, a microphone and voice recognition engine, and/or any other means of inputting data into a computing device. In the example shown in the figures, the input device 26 is an IR receiver and the audience is provided with one or more conventional IR transmitters for remotely entering data into the apparatus 10. As also shown in
For the purpose of determining if one or more members of the audience is not being identified in response to a prompt from the prompter 24, the audience change detector 12 is further provided with a compliance detector 32. As shown in
In the event such unidentified audience member(s) are detected, the compliance detector 32 adjusts a value representative of the previous number of people in the audience by a difference between the number of members identified by the audience and the number of people determined from the image(s) by the people counter 20 to avoid excessive prompting of the audience. In other words, the value indicative of the last audience count made by the people counter 20 is adjusted so that, assuming the audience composition does not change in the interim, at the next image collection and evaluation by the people counter 20, the change detector 22 will compare the audience count developed by the people counter 20 to an audience count which includes the unidentified audience member(s). Therefore, since in this example, no change in the number of audience members has occurred, the change detector 22 will not detect a change and the prompter 24 will not prompt the audience even though the unidentified audience member(s) are present. As a result, the compliance detector 32 functions to avoid excessively prompting the audience if an audience member is refusing to identify himself/herself.
In the example of
In the example of
An example apparatus 60 for implementing the apparatus 10 of
The apparatus 50 of the instant example includes a processor 54. For example, the processor 54 may be implemented by one or more IntelŪ microprocessors from the PentiumŪ family, the Itanium™ family or the XScale™ family. Of course, other processors from other families are also appropriate.
As is conventional, the processor 54 is in communication with a main memory 30 via a bus. The memory 30 stores the data developed by the apparatus 10. It also stores computer readable instructions which, when executed, cause the processor 54 to determine a number of people within the image(s) captured by the sensor 18, and to develop a prompt signal requesting the audience to identify its member(s) if a change in the number of people in the audience is visually detected
The memory 30 may include a volatile memory and a non-volatile memory. The volatile memory may be implemented by Synchronous Dynamic Random Access Memory (SDRAM), Dynamic Random Access Memory (DRAM), RAMBUS Dynamic Random Access Memory (RDRAM) and/or any other type of random access memory device. The non-volatile memory may be implemented by flash memory and/or any other desired type of memory device. Access to the main memory 30 may be controlled by a memory controller (not shown) in a conventional manner.
The memory 30 may also include one or more mass storage devices for storing software and data. Examples of such mass storage devices include floppy disk drives, hard drive disks, compact disk drives and digital versatile disk (DVD) drives.
The apparatus 50 also includes a communication block or interface circuit 56. The interface circuit 56 may be implemented by any type of well known interface standard, such as an Ethernet interface, a universal serial bus (USB), and/or a third generation input/output (3GIO) interface.
One or more input devices 56 are included in or connected to the interface circuit 56. The input device(s) permit a user to enter data and commands into the processor 54. The input device(s) can be implemented by, for example, an IR transmit/receive pair, a keyboard, a mouse, a touchscreen, a track-pad, a trackball, isopoint and/or a voice recognition system.
An output device 24 is also connected to the interface circuit 56. The output device 24 is responsive to the prompt signal output by the processor 54 to output an indication requesting the audience to identify its members. In the example of
The interface circuit 56 also includes a communication device such as a modem or network interface card to facilitate exchange of data with external computers via a network (e.g., an Ethernet connection, a digital subscriber line (DSL), a telephone line, coaxial cable, a cellular telephone system, etc.). It may also include a communication device such as an infrared decoder to receive and decode IR signals transmitted to the apparatus 60 by one or more audience members
An example software program for implementing the apparatus of
In the example of
Assuming the information presenting device is in an on state, the people prompter 24 is driven to prompt the audience to identify its member(s) (block 106). The apparatus 60 then awaits an input from the audience (block 108). When an audience input is received via, for example, the input device 26 (block 108), the processor 54 updates the database with the input data (e.g., an audience member's identity) (block 110). The time stamper 28 may record a time and date in association with the input data. The people counter 20 then increments the LAST COUNT variable to reflect the presence of the audience member that identified himself/herself (block 112).
The processor 54 then determines if a predetermined length of time (e.g., 10 seconds) has elapsed since the last input was received from the audience (block 114). If the predetermined time has not elapsed (block 114), control returns to block 108 where the processor 54 determines if another audience input has been received. If not, control again proceeds to block 114. Otherwise control advances to block 110 where the database stored in the memory 30 is updated to reflect the new audience data input. Control continues to loop through blocks 108–114 until no audience inputs are received for the predetermined length of time (block 114), at which point it is assumed that all audience members have identified themselves (although this assumption is tested at block 146 as explained below).
Assuming the predetermined length of time has elapsed without any further audience inputs (block 114), control proceeds to block 116. At block 116, the program detector 34 identifies the source of the program being presented on the information presenting device. If a change in the source has occurred (e.g., tuning changed from channel 3 to channel 11), or if a power on event just occurred (e.g., tuning changed from no tuned channel to channel 12) (block 118), the database stored in the memory 30 is updated with the new source information (block 120). As explained above, the time stamper 28 associates a time and date with the new source information. If no source change or turn on event has occurred (block 118), control skips block 120 and proceeds directly to block 122.
At block 122, the image sensor 18 is activated to capture image(s) of the audience. The captured image(s) are digitized (block 124) and passed to the people counter 20. The people counter 20 then analyzes the image(s) to determine if any person is located in the image(s) as explained below in connection with
At block 134, the change detector 22 determines if the CURRENT COUNT value (i.e., the number of persons counted in the captured image(s)) is equal to the LAST COUNT value (i.e., the number of persons counted immediately prior to the capturing of the image(s) being analyzed). If the CURRENT COUNT value and the LAST COUNT value are equal (block 134), control returns to block 116 because no audience change has occurred. Otherwise, control proceeds to block 136. Control continues to loop through blocks 116–134 until an audience count change is detected (block 134).
Assuming an audience count change has been detected (block 134), the time stamper 28 updates the database in the memory 30 with an entry indicating the time and date that an audience change occurred (block 136). It then drives the prompter 24 to prompt the audience member(s) to identify themselves(s) (block 138,
Assuming that the audience has stopped inputting data (block 144), the compliance detector 32 determines if the number of audience members identified by the inputs received from the audience is equal to the CURRENT COUNT developed from the captured image(s) by the people counter 20 (block 146). If the audience identified less audience members than the people counter 20, then the compliance detector 32 determines whether this discrepancy has occurred a predetermined number of times sequentially (e.g., three times in a row) (block 148). If not, control proceeds to block 150. Otherwise, control advances to block 156 of
Assuming for the moment that the number of audience members identified in the inputs received from the audience is equal to the number of individuals counted by the people counter 20, the compliance detector 32 sets the LAST COUNT variable equal to the CURRENT COUNT value (block 150). Setting the LAST COUNT variable in this manner ensures that only changes in the audience count result in audience prompts (see block 134). After the LAST COUNT variable is set (block 150), the CURRENT COUNT variable and the NONCOMPLIANT PERSON COUNT variable are both re-set to zero (block 152).
The program detector 34 then verifies that the information presenting device is still in an on state (block 154). If so, control returns to block 116 (
Assuming that at least one audience member refuses to identify himself/herself for the predetermined number of times sequentially (block 148), control advances to block 156 (
After re-setting the CURRENT COUNT variable to zero (block 162), the program detector 34 then verifies that the information presenting device is still in an on state (block 164). If so, control returns to block 116 (
An example people counter 20 is schematically illustrated in
The sequence of images may be digitized when the motion detector 180 receives them, or, alternatively, the motion detector 180 may include a digitizer 52 to convert the images received from the image sensor 18 into digital images. In the example of
The motion detector 180 operates on each sequential pair of images received from the image sensor 18 to detect motion occurring between the two images. More specifically, assuming a given room containing an audience is repeatedly photographed to create a series of images as explained above, then if there is no movement for a given time period, there will be no significant difference between two successive images of the room taken during the period of no movement. Thus, the binary values of the elements in the image array for a first image will be identical (or substantially identical if noise errors or the like are present) to the binary values of the corresponding elements in the image array for a second image taken immediately after the first image. If, however, there is movement between the time at which the first image is taken and the time at which the second image is taken, the binary values of the elements in the image array for the second image will be different from the binary values of the corresponding elements in the image array for the second image.
The motion detector 180 takes advantage of this fact to detect differences due to, for example, motion of audience members between successive images received from the image sensor 18 by comparing each successive pair of images on an element by element basis. In particular, the motion detector 180 develops a difference image corresponding to each pair of successively received images by subtracting the corresponding elements of one image array from the other image array. In an extremely simplified example wherein each digitized image is an array of four elements, assuming that the elements of a first received image have the following values (90, 103, 23, and 203), and the corresponding elements of a second received image have the following values (90, 103, 60 and 250), then the difference image computed by the motion detector is an array of four elements having the following values (0, 0,−37,−47). In this example, there has been motion between the first and second image and, thus, some of the values in the difference image are non-zero. The non-zero values represent points of motion. If there is no difference between successive images, all the values in the difference image corresponding to those two successive images will be zero (or substantially zero as some small differences may appear due to noise or other error).
From the foregoing, persons of ordinary skill in the art will appreciate that each difference image is typically a collection of motion points localized around center(s) of motion. In order to correlate these motion points to objects in the images, the people counter 20 is further provided with a shape. outliner 182. The shape outliner 182 employs a process such as the well known convex hull algorithm to draw shapes or blobs encompassing the motion points. As is well known by persons of ordinary skill in the art, the convex hull algorithm joins all points in a set of points that satisfy a predetermined constraint into a blob or shape. The predetermined constraint may be a requirement that all of the points in the blob or shape are separated by less than a predetermined distance. Since in this example, we are attempting to identify humans, the predetermined distance should be a distance corresponding to the size of a human being. This distance may be a settable or programmable parameter and may be set based on the sizes of the expected audience members at a given household.
Since there may not be enough data points in a difference image for the shape outliner 182 to draw meaningful shapes, the people counter 180 is further provided with an image amalgamator 184. For each image for which the people counter 20 is requested to develop a people count, the image amalgamator 184 integrates or otherwise smoothes or filters the difference images from a time interval in which the image to be analyzed is located into a single amalgamated image. For example, if the image to be analyzed occurs at time i, the image amalgamator 184 will combine the difference images from a time interval beginning at time i−k and ending at time i+c into a single image array, where k and c are preferably equal, but may be different. The difference images may be combined into an amalgamated image by summing the array corresponding to the difference images on an element by element basis and then dividing each summed element by the number of elements summed (i.e., the number of difference images). Thus, like the arrays corresponding to the difference images, the amalgamated image is an array of 8 bit binary values (i.e., values ranging from 0 to 255).
As shown in
From the foregoing, persons of ordinary skill in the art will appreciate that the motion detector 180, the image amalgamator 184 and the shape outliner 182 function to reduce the problem of counting people appearing in an image to counting blob(s) reflecting center(s) of motion within an image.
For the purpose of discriminating human blob(s) appearing within the amalgamated image from non-human blob(s) (e.g., pets, random noise, inanimate objects, etc.), the people counter 20 may optionally be further provided with a non-human filter 188. In the illustrated example, the non-human filter 188 analyzes the shape(s) drawn within the amalgamated image by the shape outliner 182 to determine if any can be eliminated from the amalgamated image as not possibly corresponding to a human being. The non-human filter 188 may employ any logical test to eliminate blob(s) from the amalgamated image. For example, the non-human filter 188 may test the location(s) of the blob(s) to determine if their location(s) identify them as not human. For instance, a blob located on the ceiling of a room can be eliminated as not human. In addition to location based tests, the non-human filter 188 may also test the size of the shape. For example, if the size of a blob is beneath a certain threshold or above a certain threshold, it may be eliminated as not reflecting a human sized object. The tests performed by the non-human filter 188 may be adjusted to suit the household being analyzed. For example, in a household with children, the non-human filter 188 may employ a lower size threshold than a household with no children. Similarly, in a household with no children, the non-human filter 188 may identify blob(s) appearing on the floor as non-human, whereas is may not be allowed to identify blob(s) on the floor as non-human based purely on a floor location if the household includes children. If the test(s) employed by the non-human filter 188 are to be tailored to the demographics of the household being analyzed, the test(s) should be adjusted at set up of the apparatus 20.
The non-human filter 188 may eliminate a blob from the amalgamated image in many different ways. For example, the binary values in the amalgamated image giving rise to the object being eliminated can be zeroed, and the revised amalgamated image fed back to the shape outliner 182 to create a new set of blob(s) in the amalgamated image excluding the blob(s) eliminated by the non-human filter 188.
For the purpose of determining if any of the blob(s) appearing in the amalgamated image (optionally, as filtered by the non-human filter 188) represents a person, the people counter 20 is further provided with a blob discriminator 190. Were one to simply count the number of blobs appearing in the amalgamated image (optionally as filtered by the non-human filter 188), false people counts might result in certain instances. For example, if two people are located in an audience, but only one of those people moves during a time period being analyzed, only one blob will appear in the amalgamated image, and simply counting blobs without further refinement would result in an undercount. By way of another example, if two audience members move in a symmetrical fashion for a given period of time, they could potentially appear as a single blob in the amalgamated image. Simply counting blobs in this scenario will again result in an undercount. The blob discriminator 190 solves this potential problem by ensuring only blob(s) that exhibit persistent motion over a time period of interest are counted as persons.
To perform the persistent motion test, the blob discriminator 190 does not develop a count of the blobs appearing in every amalgamated image. Instead, a number of sequential amalgamated images are analyzed over a period of time. In particular, for each amalgamated image, the blob(s) contained therein are represented by symbols in a histogram. Although a blob can appear only once in any given amalgamated image, if the blob exhibits persistent motion, it will appear in multiple different amalgamated images. For every time a blob appears in an amalgamated image and meets the convex bull criteria, a symbol is added to the histogram. Therefore, the histogram tracks the number of times each blob exhibits motion over a period of time. After that period of time, the histogram is analyzed and only those blobs that have exhibited sufficient persistence of motion as indicated by the number of times a symbol corresponding to that blob appears in the histogram, are identified as persons.
An example blob discriminator 190 is shown in
In order to add a symbol which is representative of the center of gravity of the blob to the histogram, the blob discriminator 190 is further provided with a center comparator 194. The center comparator 194 serves a gravitation function. In particular, whenever the center locator 192 computes a center of gravity of a blob, the center comparator 194 compares the newly computed center of gravity to the existing centers of gravity already appearing in the histogram. If the newly computed center of gravity is the same as, or falls within a predetermined distance of, a center of gravity already represented in the histogram, it is assumed that the newly computed center represents the same object as the existing center. As a result, a symbol representative of the newly computed center is added to the symbol representing the existing center in the histogram. Preferably, every symbol added to the histogram has the same size. Therefore, when a symbol is added to one or more existing symbols in the histogram, the existing symbol “grows” in size.
Persons of ordinary skill in the art will readily appreciate that a histogram such as that described above may be implemented in many different ways. For example, it may be implemented graphically wherein symbol(s) of the same size are placed at the X-axis location of their corresponding blob(s). If two or more symbols have substantially the same X-axis location (thereby exhibiting some level of persistent motion of their corresponding object), they are stacked vertically. Alternatively, a horizontal growth metric may be used. Alternatively or additionally, the histogram could be implemented by a set of counters wherein each counter in the set corresponds to an X-axis location within an amalgamated image. If a blob having a center of gravity corresponding to the X-axis location of a given counter is identified in an amalgamated image, the corresponding counter is incremented. Therefore, the larger the number of times a blob appears in a series of amalgamated images, the larger the value in the corresponding counter becomes.
To determine whether any symbol in the histogram has exhibited sufficient persistent motion to be counted as a person in the audience, the blob discriminator 190 is further provided with a threshold counter 198. The threshold counter 198 compares the number of times each center of gravity is represented in the histogram to a predetermined threshold. This can be done, for example, by comparing the size of the symbol to the predetermined threshold. If any symbol in the histogram has a size greater than the threshold, it is counted as a person. Thus, in the example of
In the people counter 20 of
The people counts developed at the motion markers can be extrapolated to the periods occurring between motion markers. This extrapolation is possible because there is relatively little motion between the motion markers. Someone entering or leaving the audience room is associated with a significant amount of motion. Since motion markers are set when such a significant amount of motion occurs, no significant amount of motion occurs between motion markers, and it can, thus, be safely assumed that no one has left or entered the room in the time between motion markers. Therefore, it can be safely assumed that the audience composition has not changed in the time between motion markers. By way of an example, if at a first motion marker the people counter 20 determines there are 2 people in the audience, and at the next motion marker the people counter determines there are three people in the room, then, because no motion indicating a person has entered or exited the room is detected prior to the second motion marker, the people count for the entire period from the first motion marker to the second motion marker is two people.
When sufficient data has been developed around a motion marker (e.g., when enough amalgamated images prior to and after a motion marker have been analyzed for the corresponding histogram developed from those amalgamated images to have meaning, the threshold counter 198 is activated. As explained above, the threshold counter 198 determines whether any blob represented in the histogram has exhibited sufficient persistence of motion to be counted as a person. Any such blob is counted as a person and the person count so developed is output by the people counter 20.
To prevent noise and false motions from cluttering the histogram, the blob discriminator 190 is further provided with a false motion filter 202. The false motion detector 202 of the illustrated example reviews the symbols recorded in the histogram as the histogram is being developed. If any symbol does not grow for a predetermined amount of time (e.g., three minutes), the symbol is assumed to be noise or other false motion and is eliminated from the histogram. In this way, erroneous entries due to noise, random movement or erroneous consolidation of two or more blobs into one blob are not allowed to grow.
An example software program for implementing the apparatus 20 of
The program of
The motion detector 180 then computes the difference image between the most recently captured image and the immediately proceeding image stored in memory (block 226). As discussed above, the difference image is calculated by subtracting the elements of the most recently captured image array from the corresponding elements of the most recently stored image array in accordance with the conventional rules of linear algebra.
Once the difference image is calculated (block 226), the energy detector 200 calculates the energy value associated with the difference image (block 228). As explained above, this energy value is computed by squaring the array of the difference image and then summing all of the values contained in the array generated by the squaring operation. The energy value is stored in memory 30 for later use as explained below.
Because many of the calculations performed by the people counter 20 require data corresponding to images taken before and after a motion marker, it is necessary to have a running sequence of pictures to operate upon. Therefore, before creating any amalgamated images, the people counter 20 creates a buffer of captured images and difference images. Thus, at block 230, the people counter 20 determines whether the desired buffer has been created. If not, control loops back to block 220 via block 232. At block 232 a captured image counter i is incremented. Control continues to loop through blocks 220–232 until the desired buffer of captured images and difference images corresponding to those captured images has been created (block 230).
Assuming the desired buffer is in place (block 230), control advances to block 234. At block 234 a number of counters are initialized. For example, an amalgamation counter A is set to equal the image counter i less the buffer size. A histogram loop counter B is set to equal the amalgamation counter less a delay value sufficient to ensure all needed amalgamation image arrays have been computed prior to initiating population of a histogram. A range counter K is set to equal the amalgamation counter A−Z, a variable setting the earliest difference image to be used in creating an amalgamated image corresponding to time A. A range counter M is set to equal the histogram loop counter B−P, a variable setting the earliest amalgamated image to be used in creating a histogram corresponding to a motion marker occurring at time B. A threshold T is set equal to the amalgamation counter A+E, a variable setting the latest difference image to be used in creating an amalgamated image corresponding to time A. A second threshold U is set to equal the histogram loop counter B+F, a variable setting the latest amalgamated image to be used in creating a histogram corresponding to a motion marker occurring at time B. Additionally, the amalgamation array SA for time A is cleared to an empty set. Persons of ordinary skill in the art will appreciate that the variables Z and E may optionally be identical. Similarly, the variables P and F may optionally be identical.
Once the variables are initialized as explained above (block 234), the image amalgamator 184 sums the amalgamation array SA with the difference image array associated with time K on an element by element basis (block 236). The counter K is then incremented (block 238). If the counter K has not surpassed the threshold T (block 240), control returns to block 236 where the image amalgamator 184 adds the next difference image array to the amalgamated image array. Control continues to loop through blocks 236–240 until the counter K equals or exceeds the threshold T (block 240).
When the compilation of the amalgamated image array is completed (block 240), the image amalgamator 184 converts the amalgamated image array SA into a binary image (block 242). Converting the amalgamated image array to a binary image can be accomplished by, for example, dividing each element in the amalgamated image array by the number of difference images used to form the amalgamated image array (e.g., by (Z+E)).
The energy detector 200 then determines whether the energy value associated with time B is greater than an energy threshold X (i.e., whether a motion marker is set at time B) (block 244). The energy threshold X is a value that indicates the amount of movement energy that is required in a difference image to suggest that an audience composition change is occurring. If a motion marker is set at time B, then control advances to block 250. Otherwise, the people counter routine determines whether it has been executing for too long of a time (block 246). If so, the people counter routine terminates and control advances to block 134 of
Assuming it is not time to exit the people counter routine to check for source changes or a turn-off event (block 246), control advances to block 248. At block 248 the captured image counter i is incremented. Control then returns to block 220 (
Assuming for purposes of discussion that a motion marker is located at time B (block 244), control enters a loop wherein a histogram corresponding to the time period beginning at time M (i.e., time (B−P)) and ending at time U (i.e., time (B+F)) is populated. In particular, at block 250, the shape outliner 182 executes the convex hull process on the points appearing in the amalgamated image array SM corresponding to time M. As explained above, if any points are presenting the amalgamated image array SM, the execution of the convex hull process draws one or more blob(s) in the amalgamated image array SM.
Once the blob(s) (if any) are drawn, the non-human filter 188 performs one or more logic test(s) on the blob(s) to attempt to eliminate non-human blob(s) from the amalgamated image array SM (block 252). As explained above, many different-logic tests may be used for this purpose including, by way of examples, not limitations, a location test and/or a size test.
Once the non-human filter 188 has completed execution, the center locator 192 calculates the center of gravity of each remaining blob (if any) in the amalgamated image array SM (block 254). As explained above, this calculation may be performed by averaging the X-axis values for each point in the blob in question.
Irrespective of how the center(s) of the blob(s) are identified, once the centers are calculated, the center comparator 192 attempts to record the blob(s) in the histogram. In particular, the center comparator 192 determines if the center of a first one of the blob(s) (if any) in the amalgamated image SM is located within a predetermined distance Y of a center of an object already recorded in the histogram (block 256). The predetermined distance is preferably selected to correspond to the expected size of a person along the x-axis of an image (e.g., 40 pixels). As explained above, this test is performed to ensure that slight differences in the centers of blobs do not cause the same blob to be identified as two different blobs in different amalgamated images. If the center of the blob under consideration is within Y distance of a center already existing in the histogram (block 256), a symbol representative of the blob under consideration is added to the symbol representing the already existing center in the histogram (block 258,
Irrespective of whether control passes through block 258 or 260, when control reaches block 262, the center comparator 194 determines if there are more blobs to analyze within the amalgamated image SM under examination. If so, control returns to block 256 (
At block 264, the range counter M is incremented. The blob discriminator 196 then determines whether the loop counter M is equal to or greater than the threshold U (block 266). If not, then all of the amalgamated images to be represented in the histogram have not yet been analyzed, and control advances to block 268. Otherwise, the histogram is complete and control advances to block 272.
Assuming for purposes of discussion that the histogram is not yet fully populated (block 266), the false motion filter 202 examines the histogram to determine if any symbols in the histogram have failed to grow within a predetermined time period (e.g., 3 minutes)(block 268). If any such inactive symbols exist (block 268), the false motion filter 202 assumes these inactive symbols are not representative of people and removes them from the histogram (block 270). Control then returns to block 250 (
Control continues to loop through blocks 250–270 until the loop counter M becomes equal to or greater than the threshold U (block 266). The histogram is then complete and ready for analysis. Accordingly, the histogram is latched and stored.
The threshold counter 198 then begins analyzing each symbol representative of a blob center appearing in the histogram (block 272). If a symbol being examined exceeds a predetermined threshold (block 272), the threshold counter 198 identifies the symbol as representative of a person. Accordingly, the CURRENT COUNT variable is incremented (block 274). If the symbol being examined does not exceed the predetermined threshold (block 272), the threshold counter 198 concludes that the symbol represents something other than a person and the CURRENT COUNT variable is, therefore, not incremented (block 272). The threshold counter 198 then determines if every symbol in the histogram has been analyzed (block 276). If not, control returns to block 272. Control continues to loop through blocks 272–276 until every symbol in the histogram has been identified as human or non-human and the human symbols have been counted (block 276). Once this process is completed (block 276), the histogram is cleared for the next round of analysis (block 278). The people counter routine then terminates. In the example of
To provide further illustration of the operation of the people counter 20 discussed in connection with
As shown in
The third amalgamated image contains only one blob. As shown in
As also shown in
In the final state of the histogram reflected in
Although certain example methods and apparatus constructed in accordance with the teachings of the invention have been described herein, the scope of coverage of this patent is not limited thereto. On the contrary, this patent covers all embodiments of the teachings of the invention fairly falling within the scope of the appended claims either literally or under the doctrine of equivalents.
|Cited Patent||Filing date||Publication date||Applicant||Title|
|US3805238||May 9, 1972||Apr 16, 1974||Rothfjell R||Method for identifying individuals using selected characteristic body curves|
|US4468807||Jul 13, 1981||Aug 28, 1984||Palmguard, Inc.||Method for analyzing stored image details|
|US4611347||Sep 24, 1984||Sep 9, 1986||At&T Bell Laboratories||Video recognition system|
|US4644509 *||Jan 23, 1986||Feb 17, 1987||A. C. Nielsen Company||Ultrasonic audience measurement system and method|
|US4769697||Dec 17, 1986||Sep 6, 1988||R. D. Percy & Company||Passive television audience measuring systems|
|US4843631||Dec 22, 1986||Jun 27, 1989||Dietmar Steinpichler||Pattern recognition process|
|US4849737||Nov 19, 1987||Jul 18, 1989||Matsushita Electric Works, Ltd.||Person-number detecting system|
|US4858000||Sep 14, 1988||Aug 15, 1989||A. C. Nielsen Company||Image recognition audience measurement system and method|
|US4993049||Sep 28, 1988||Feb 12, 1991||Cupps Halbert D||Electronic management system employing radar type infrared emitter and sensor combined with counter|
|US5031228 *||Sep 14, 1988||Jul 9, 1991||A. C. Nielsen Company||Image recognition system and method|
|US5063603||Nov 6, 1989||Nov 5, 1991||David Sarnoff Research Center, Inc.||Dynamic method for recognizing objects and image processing system therefor|
|US5067160||Mar 24, 1989||Nov 19, 1991||Canon Kabushiki Kaisha||Motion-pattern recognition apparatus|
|US5097328||Oct 16, 1990||Mar 17, 1992||Boyette Robert B||Apparatus and a method for sensing events from a remote location|
|US5099324||Jun 27, 1990||Mar 24, 1992||Kabushiki Kaisha Toshiba||Apparatus for extracting/combining change region in image corresponding to moving object|
|US5121201||Aug 14, 1990||Jun 9, 1992||Daido Denki Kogyo Kabushiki Kaisha||Method and apparatus for detecting the number of persons|
|US5164992||Nov 1, 1990||Nov 17, 1992||Massachusetts Institute Of Technology||Face recognition system|
|US5229764||Jun 20, 1991||Jul 20, 1993||Matchett Noel D||Continuous biometric authentication matrix|
|US5384716||Sep 22, 1993||Jan 24, 1995||Matsushita Electric Industrial Co., Ltd.||Occupant condition determining apparatus|
|US5412738||Aug 10, 1993||May 2, 1995||Istituto Trentino Di Cultura||Recognition system, particularly for recognising people|
|US5481622 *||Mar 1, 1994||Jan 2, 1996||Rensselaer Polytechnic Institute||Eye tracking apparatus and method employing grayscale threshold values|
|US5550928||Dec 15, 1992||Aug 27, 1996||A.C. Nielsen Company||Audience measurement system and method|
|US5629752||Oct 27, 1995||May 13, 1997||Fuji Photo Film Co., Ltd.||Method of determining an exposure amount using optical recognition of facial features|
|US5675663||Mar 21, 1996||Oct 7, 1997||Honda Giken Kogyo Kabushiki Kaisha||Artificial visual system and method for image recognition|
|US5715325||Aug 30, 1995||Feb 3, 1998||Siemens Corporate Research, Inc.||Apparatus and method for detecting a face in a video image|
|US5719951||Jul 17, 1991||Feb 17, 1998||British Telecommunications Public Limited Company||Normalized image feature processing|
|US5771307||Dec 21, 1995||Jun 23, 1998||Nielsen Media Research, Inc.||Audience measurement system and method|
|US5781650||Aug 28, 1997||Jul 14, 1998||University Of Central Florida||Automatic feature detection and age classification of human faces in digital images|
|US5793409||Dec 18, 1992||Aug 11, 1998||Kabushikigaisha Shogakuikueisha Kyoikukenkyusho||Apparatus for grasping TV viewing condition in household|
|US5801763||Dec 20, 1995||Sep 1, 1998||Mitsubishi Denki Kabushiki Kaisha||Face image taking device|
|US5805745||Jun 26, 1995||Sep 8, 1998||Lucent Technologies Inc.||Method for locating a subject's lips in a facial image|
|US5835616||Jun 17, 1994||Nov 10, 1998||University Of Central Florida||Face detection using templates|
|US5850470||Aug 30, 1995||Dec 15, 1998||Siemens Corporate Research, Inc.||Neural network for locating and recognizing a deformable object|
|US5859921 *||Feb 28, 1996||Jan 12, 1999||Mitsubishi Denki Kabushiki Kaisha||Apparatus for processing an image of a face|
|US5864630||Nov 20, 1996||Jan 26, 1999||At&T Corp||Multi-modal method for locating objects in images|
|US5878156||Feb 12, 1996||Mar 2, 1999||Mitsubishi Denki Kabushiki Kaisha||Detection of the open/closed state of eyes based on analysis of relation between eye and eyebrow images in input face images|
|US5892837||Aug 29, 1997||Apr 6, 1999||Eastman Kodak Company||Computer program product for locating objects in an image|
|US5901244||Dec 12, 1996||May 4, 1999||Matsushita Electric Industrial Co., Ltd.||Feature extraction system and face image recognition system|
|US5920641||Sep 6, 1995||Jul 6, 1999||Siemens Nixdorf Informationssysteme Aktiengesellschaft||Method for reconstructing linear structures present in raster form|
|US5963670||Feb 12, 1996||Oct 5, 1999||Massachusetts Institute Of Technology||Method and apparatus for classifying and identifying images|
|US5978507||Jan 28, 1993||Nov 2, 1999||British Telecommunications Public Limited Company||Method of forming a template of an image of an object for use in the recognition of the object|
|US5987154||Nov 22, 1996||Nov 16, 1999||Lucent Technologies Inc.||Method and means for detecting people in image sequences|
|US6047134 *||Oct 21, 1997||Apr 4, 2000||Canon Kabushiki Kaisha||Image shake detecting device|
|US6055323||Jan 5, 1998||Apr 25, 2000||Mitsubishi Denki Kabushiki Kaisha||Face image processing system|
|US6144797 *||Sep 2, 1997||Nov 7, 2000||Sensormatic Electronics Corporation||Intelligent video information management system performing multiple functions in parallel|
|US6332033||Aug 14, 2000||Dec 18, 2001||Sharp Laboratories Of America, Inc.||System for detecting skin-tone regions within an image|
|US6332038||Apr 6, 1999||Dec 18, 2001||Sharp Kabushiki Kaisha||Image processing device|
|US6363159||Nov 17, 1999||Mar 26, 2002||Digimarc Corporation||Consumer audio appliance responsive to watermark data|
|US6507391 *||Aug 24, 2001||Jan 14, 2003||Amnis Corporation||Measuring the velocity of small moving objects such as cells|
|US6625316 *||Jun 1, 1999||Sep 23, 2003||Canon Kabushiki Kaisha||Image processing apparatus and method, and image processing system|
|US6940545||Feb 28, 2000||Sep 6, 2005||Eastman Kodak Company||Face detecting camera and method|
|US6944319||Mar 27, 2000||Sep 13, 2005||Microsoft Corporation||Pose-invariant face recognition system and process|
|US20030033600||Jul 29, 2002||Feb 13, 2003||Cliff David Trevor||Monitoring of crowd response to performances|
|US20050198661||Jan 21, 2005||Sep 8, 2005||Andrew Collins||Display|
|WO1999027668A1||Jun 9, 1998||Jun 3, 1999||Nielsen Media Res Inc||Voice recognition unit for audience measurement system|
|WO2004054255A1||Dec 11, 2002||Jun 24, 2004||Nielsen Media Res Inc||Detecting a composition of an audience|
|1||"1.6.2 Convex Hull," http://www.cs.sunysb.edu/~algorith/files/convex-hull.shtml, downloaded Nov. 6, 2002.|
|2||"Convex Hull", http://www.cse.unsw.edu.au/~lambert/java/3d/ConvexHull.html, download, Nov. 6, 2002.|
|3||"What is convex hull? What is the convex hull problem?," http://www.ifor.math.ethz.ch/~fukuda/polyfaq/node 13.html, downloaded Nov. 6, 2002.|
|4||Duda and Hart, "Bayes Decision Theory", Pattern Classification & Scene Analysis, pp. 10-43, Wiley Publishers, 1973.|
|5||IBM Exploratory Computer Vision Group, "Infrared Person Tracking," Jun. 12, 2002, 2 pages, www.research.ibm.com/ecvg/misc/footprint.html.|
|6||International Preliminary Examination Report corresponding to International Application No. PCT/US02/39619, Jun. 1, 2004.|
|7||International Preliminary Examination Report corresponding to International Application No. PCT/US2002/039625, Apr. 27, 2004, 4 pages.|
|8||International Search Report corresponding to International Application No. PCT/US02/39619, Mar. 11, 2003.|
|9||International Search Report corresponding to International Application No. PCT/US2002/039625, Jun. 20, 2003, 5 pages.|
|10||New Scientist Print Edition, "Hot Shopping," Mar. 18, 2000, 1 page, www.newscientist.com/article.ns?id=mg16522301.700&print=true.|
|11||Qing et al. "Histogram Based Fuzzy C-Mean Algorithm For Image Segmentation," Pattern Recognition, 1992. vol. III. Conference C: Image, Speech and Signal Analysis, Proceedigns., 11th IAPR International Conference, Aug. 30-Sep. 2, 1992.|
|12||Ramaswamy et al., "Detecting a composition of an audience," U.S. Appl. No. 10/538,692, filed Nov. 12, 2002.|
|13||Wolfram, Stephen, "Mathematica A System for Doing Mathematics by Computer", 2d Ed., Addison-Wesley Publishing Company, Inc., 1991, pp. 665-666.|
|14||Written Opinion of the International Preliminary Examining Authority corresponding to International Application No. PCT/US02/39619, Apr. 20, 2004.|
|15||Written Opinion of the International Preliminary Examining Authority corresponding to International Application No. PCT/US02/39619, Aug. 22, 2003.|
|Citing Patent||Filing date||Publication date||Applicant||Title|
|US7609853 *||Dec 11, 2002||Oct 27, 2009||The Nielsen Company (Us), Llc||Detecting a composition of an audience|
|US7991193||Jul 30, 2007||Aug 2, 2011||International Business Machines Corporation||Automated learning for people counting systems|
|US8194923||Jul 31, 2009||Jun 5, 2012||The Nielsen Company (Us), Llc||Methods and apparatus for detecting a composition of an audience of an information presenting device|
|US8285713||Sep 8, 2008||Oct 9, 2012||International Business Machines Corporation||Image search using face detection|
|US8334906 *||May 4, 2007||Dec 18, 2012||Objectvideo, Inc.||Video imagery-based sensor|
|US8411963||Aug 7, 2009||Apr 2, 2013||The Nielsen Company (U.S.), Llc||Methods and apparatus to count persons in a monitored environment|
|US8620088 *||Aug 31, 2011||Dec 31, 2013||The Nielsen Company (Us), Llc||Methods and apparatus to count people in images|
|US8660308||Apr 30, 2012||Feb 25, 2014||The Nielsen Company (Us), Llc||Methods and apparatus for detecting a composition of an audience of an information presenting device|
|US8660895||Jun 14, 2007||Feb 25, 2014||Videomining Corporation||Method and system for rating of out-of-home digital media network based on automatic measurement|
|US8677385||Sep 21, 2010||Mar 18, 2014||The Nielsen Company (Us), Llc||Methods, apparatus, and systems to collect audience measurement data|
|US8678624||Sep 30, 2011||Mar 25, 2014||The Nielsen Company (Us), Llc||Apparatus and methods for illuminating display panels of electronic devices|
|US8737745||Mar 27, 2012||May 27, 2014||The Nielsen Company (Us), Llc||Scene-based people metering for audience measurement|
|US8824740||Dec 23, 2013||Sep 2, 2014||The Nielsen Company (Us), Llc||Methods and apparatus for detecting a composition of an audience of an information presenting device|
|US9016906||Dec 24, 2013||Apr 28, 2015||The Nielsen Company (Us), Llc||Apparatus and methods for illuminating display panels of electronic devices|
|US9055334||Dec 30, 2013||Jun 9, 2015||The Nielsen Company (Us), Llc||Methods, apparatus, and systems to collect audience measurement data|
|US9161084||Oct 29, 2013||Oct 13, 2015||Videomining Corporation||Method and system for media audience measurement by viewership extrapolation based on site, display, and crowd characterization|
|US9174351||Jun 20, 2012||Nov 3, 2015||May Patents Ltd.||Electric shaver with imaging capability|
|US9185456||Mar 27, 2012||Nov 10, 2015||The Nielsen Company (Us), Llc||Hybrid active and passive people metering for audience measurement|
|US20060200841 *||Dec 11, 2002||Sep 7, 2006||Arun Ramaswamy||Detecting a composition of an audience|
|US20070195995 *||Feb 16, 2007||Aug 23, 2007||Seiko Epson Corporation||Calculation of the number of images representing an object|
|US20070285510 *||May 4, 2007||Dec 13, 2007||Object Video, Inc.||Intelligent imagery-based sensor|
|US20090158309 *||Dec 12, 2007||Jun 18, 2009||Hankyu Moon||Method and system for media audience measurement and spatial extrapolation based on site, display, crowd, and viewership characterization|
|US20090290756 *||Nov 26, 2009||Arun Ramaswamy||Methods and apparatus for detecting a composition of an audience of an information presenting device|
|US20100027875 *||Jul 30, 2007||Feb 4, 2010||Arun Hampapur||Automated learning for people counting systems|
|US20110317017 *||Aug 20, 2010||Dec 29, 2011||Olympus Corporation||Predictive duty cycle adaptation scheme for event-driven wireless sensor networks|
|US20120150586 *||Jun 14, 2012||Scenetap Llc||Apparatus and method to record customer demographics in a venue or similar facility using cameras|
|US20130051677 *||Feb 28, 2013||Morris Lee||Methods and apparatus to count people in images|
|U.S. Classification||382/103, 382/168, 348/169|
|International Classification||G06K9/00, H04H1/00, H04H60/33|
|Cooperative Classification||G06K9/00778, G06K9/00369, G06T7/2053, H04H60/33|
|European Classification||G06K9/00V4C, G06T7/20D, G06K9/00H1, H04H60/33|
|Apr 28, 2003||AS||Assignment|
Owner name: NIELSEN MEDIA RESEARCH, INC., NEW YORK
Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNORS:RAMASWAMY, ARUN;NELSON, DANIEL J.;SRINIVASAN, VENUGOPAL;REEL/FRAME:014003/0840
Effective date: 20021210
|Sep 7, 2006||AS||Assignment|
Owner name: CITIBANK, N.A., AS COLLATERAL AGENT,NEW YORK
Free format text: SECURITY AGREEMENT;ASSIGNORS:NIELSEN MEDIA RESEARCH, INC.;AC NIELSEN (US), INC.;BROADCAST DATA SYSTEMS, LLC;AND OTHERS;REEL/FRAME:018207/0607
Effective date: 20060809
|Aug 12, 2008||CC||Certificate of correction|
|Jun 29, 2009||AS||Assignment|
Owner name: NIELSEN COMPANY (US), LLC, THE, ILLINOIS
Free format text: MERGER;ASSIGNOR:NIELSEN MEDIA RESEARCH, LLC (FORMERLY KNOWN AS NIELSEN MEDIA RESEARCH, INC.) A DELAWARE CORPORATION;REEL/FRAME:022892/0544
Effective date: 20081001
|Oct 12, 2010||FPAY||Fee payment|
Year of fee payment: 4
|Oct 10, 2014||FPAY||Fee payment|
Year of fee payment: 8